Business analytics software vendor Sisense Inc. has introduced a new BI tool that aims to use deep learning in...
big data analytics to help enterprises make use of their unstructured data.
Called Sisense Hunch Data Cognition Engine (Sisense Hunch), the vendor said the system will be able to save organizations time and money by automatically extracting insights from big data. But that could possibly be at the cost of scope and accuracy.
Neural networks in big data
Essentially, Sisense Hunch is "a general purpose analytics engine" powered by a "multilevel neural network," said Amir Orad, Sisense CEO.
"It will answer any question you ask it," Orad said, as long as the query is asked in the SQL programming language.
Sisense, founded in 2004, sells a self-contained integrated analytics and BI platform that covers the entire workflow. The company has long touted its products' ease of use, fast implementation time and relatively fast big data analysis times. Hunch appears to fit in as a complementary addition to the vendor's product suite.
Hunch, sold with subscription-based pricing based on an organization's data sizes and types, uses deep learning in big data analytics to automatically explore big data and discover useful insights.
Getting both questions and answers
Boris Evelsonanalyst, Forrester
Sisense intends Hunch, formally released Oct. 23 and commercially available now, to take some of the conjecture out of the useful information that actually resides in an organization's data, said Boris Evelson, a Forrester analyst.
With Sisense Hunch, "rather than just opening up the database to an infinite number of questions, let's run neural network processes through that data set and let machine learning tell us what questions can be answered," Evelson said.
Hunch then runs the questions and creates a database populated by those questions, as well as the answers.
This new database, containing only the questions and answers obtained from the original data, is lightweight, and can fit on small devices such as smartphones, wearables and sensors, enabling big data analytics on IoT devices. The condensed data packet enables a less compute-intensive and faster way to query the data and find insights.
It also creates an extra layer of security and privacy, Evelson noted.
"You can analyze your questions and answers, but you can't really see the underlying data," he said. So, if the system is properly trained, private details, for example, won't show up -- only the gist of what was originally presented will.
As Orad put it, the neural network, like a human brain, "learns it and forgets the details."
By using deep learning in big data analytics, organizations could be able to save time, manpower and costs. Large organizations tend to store about 80 % of their data in data lakes.
Orad said the software "learns the data by deployment," and may only take a few days for an organization to set up and train.
A drag-and-drop interface also means Sisense Hunch can be used by citizen data scientists, Orad said.
However, there is a big potential drawback.
Compromise and concession
By enabling automated condensing with deep learning in big data analytics, rather than relying on, say, a large, slow data warehouse, users sacrifice a degree of accuracy for faster results.
According to Sisense, Hunch can provide results with 99% accuracy. And, even then, that number is likely only possible if users work with a limited scope and properly train the deep learning in the big data analytics tool.
Orad acknowledged that the "difference between 99 and 100 is endless." But, he added, for users, sacrificing that 1% might save on "tens of thousands of resources."
"The alternative is to go to a big IT team and ask it to build a big data warehouse," he said. "Our process is much more intuitive and the time to insight, the time to value, is much more accelerated."
Evelson said that, likely, for the majority of use cases, "99% is perfectly fine."
He said he believes Sisense Hunch could provide large enterprises, which tend to have massive amounts of raw data, with a means to start to make use of that data.
For smaller or medium-sized businesses, which might possess considerably less unstructured data, Evelson said the usefulness of the product would ultimately come down to the price.